In this project I created a simple CNN model with data augmentation to classify sign language hand digits [dataset source]. With about 20 epochs, the model achieves a 98% accuracy on the test set (I set 30 epochs, and used an "early stopping" callback which has stopped the model just when it started converging).
A few prediction examples (from the test set) can be seen in the following snapshot:
For a full walkthrough of the code, please head over to the sign_language_digits.ipynb notebook in the root directory of this repo.